Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Erdos 1989 additive bases many representations
P. Erdős, M. B. Nathanson: Additive bases with many representations, Acta Arith. 52 (1989), no. 4, 399--406. DOI 10.4064/aa-52-4-399-406; MR 91e:11015; Zentralblatt 692.10045.
The paper addresses whether an asymptotic basis A of order 2 with r_A(n) -> infinity must contain a minimal asymptotic basis, the earlier Erdos-Nathanson result needing r_A(n) >= c log n with c > 1/log(4/3). Writing S_A(n) = {a in A : n - a in A}, Lemma 1 says that if |S_A(n) n S_A(u)| < |S_A(n)|/2 then n lies in 2(A \ S_A(u)). Theorem 1 uses this to prove that if r_A(n) -> infinity and there is delta > 0 with |S_A(n) n S_A(m)| < (1/2 - delta)|S_A(n)| for all large distinct m, n, then A contains a minimal asymptotic basis of order 2; the construction is an explicit greedy induction that repeatedly deletes a solution set S_A(u_k) while keeping the sumset 2A unchanged and making u_k uniquely representable. In the other direction Theorem 2 (p. 404) proves that for every integer t there is a set A with r_A(n) >= t for all large n such that, for every subset S of A, A \ S is an asymptotic basis of order 2 if and only if S is finite; in particular A contains no minimal asymptotic basis of order 2. The proof refines a construction the authors had used to build a basis A for which A \ S is a basis exactly when A n S is finite. Theorem 5 (p. 406), stated as a simple corollary of Theorem 2, gives for every integer t an asymptotic basis A of order 2 with r(n) >= t for all large n that is not the union of two disjoint asymptotic bases of order 2, in contrast with the authors' earlier partition theorem under r_A(n) >= c log n, c > 1/log(4/3). Together these results say the small-overlap hypothesis, not merely a bounded-below representation count, is what forces minimality; Theorem 2 bears on problem 868 (minimal subbases) and Theorem 5 on problem 871 (partition into two bases), both of which ask about r_A(n) -> infinity rather than r_A(n) >= t.
Source: https://users.renyi.hu/~p_erdos/1989-03.pdf. The file's text layer carries no copyright or license line; the journal's record offers the PDF under the download link "Pobierz zgodnie z CC-BY", rendered "Free download under CC-BY license" on the English site, and names no version or URL for it (https://www.impan.pl/get/doi/10.4064/aa-52-4-399-406, read 2026-10-02): the Creative Commons Attribution license, with no version stated.
Results to transcribe.
- Lemma 1: If |S_A(n) n S_A(u)| < (1/2)|S_A(n)| then n is in 2(A \ S_A(u)); i.e. deleting the solution set of u does not destroy n.
- Theorem 1: If r_A(n) -> infinity and |S_A(n) n S_A(m)| < (1/2 - delta)|S_A(n)| for all large m != n, then the asymptotic basis A of order 2 contains a minimal asymptotic basis of order 2.
- Theorem 2: For every integer t there is a set A of nonnegative integers with r_A(n) >= t for all large n such that, for every subset S of A, A \ S is an asymptotic basis of order 2 if and only if S is finite; in particular A contains no minimal asymptotic basis of order 2.
- Theorem 5: For every integer t there is an asymptotic basis A of order 2 with r(n) >= t for all large n that is not the union of two disjoint asymptotic bases of order 2.
- Earlier threshold recalled: Erdos-Nathanson: r_A(n) >= c log n with c > 1/log(4/3) already forces a minimal asymptotic basis; whether r_A(n) -> infinity suffices is stated as the open problem.